Route 121: the 2023 Oakland A’s
Probability that the 2023 Oakland A’s Eclipse the 1962 New York Mets’ Loss Total
Growing up a Mets fan, I was told early on of the historically bad inaugural season our lovable losers had. In 161 games (40-120-1, 0.250), the Mets won just 40, lost 120, and mercifully tied their fellow expansion Houston Colt .45s one day after losing both ends of a doubleheader -- each on walk-offs (two of twelve walk-off losses that year). They had three separate losing streaks of at least 10 games, including a 17-game stretch that put them 24 games behind on June 6. Barring the 1899 Cleveland Spiders, they’re basically the worst team you can imagine.
Well, the 2023 A’s are giving the ‘62 Mets a run for their money. This year’s A’s aren’t lovable losers, they’re the recipients of our pity and disgust. They’re a team built to lose, and they’re doing a damn good job of it. Amidst the turmoil of an empty ballpark in shambles, the looming move to Las Vegas, and the recent memory of 97-win teams, the A’s have started the season 12-47, good for a 0.203 clip. Since 1933, that’s the worst 59-game start to a season. The Mets are 23rd on that list.
Extrapolated out over the course of a 162 game season, their current pace would leave the 2023 A’s 33-129. Clearly, though, we can’t just multiply out. There have been plenty of teams that have started a season off poorly, most recently the 2016 Atlanta Braves, who went 17-42 to start the year (just one game better than the ‘62 Mets), but “only” lost 93 games. Some regression to the mean is expected here.
So, inspired by the work of Professor Michael Richmond and Dan Szymborski, I decided to find out for myself just how likely it is that the A’s can remove the burden of the ‘62 team from my Mets fandom.
Richmond’s work from 2007 demonstrates the flaws with extrapolating out a team’s current win percentage to find their final record, especially this early in the season:
Instead, Richmond proposes a simple linear model which consists of an A and B term, both of which fluctuate depending on the number of games played. He shows it as the following:
Final Winning Percentage = A + B*(Current Winning Percentage)
Szymborski, on the other hand, uses his Fangraphs ZiPS projection model to determine the probability that the A’s lose more than 120 games. Through 48 games, he pinned that probability to be ~3.6%. The A’s have gone 2-9 since then.
My goal was to combine these two methods: develop my own set of A and B parameters for each game, and then parlay that into a simulation to determine how often the A’s eclipse 120 losses.
To do this, I first downloaded every game that every team has played since 1962. (I originally downloaded every game since 1903, but started with the modern expansion era in 1962 -- this was also the first year in which every team played 162 games). For each n in [1, 2, 3, … 162], I needed to correlate a team’s current winning percentage with their final winning percentage. For instance, through 59 games, that data looked something like this:
Overall, the set of A and B values I calculated for each game of the season looked like this:
Similar to what Professor Richmond had produced for AL teams 1961-2006:
I then produced a plot demonstrating the correlation between the current and final winning percentage:
Using the simple linear model for the current number of games played and the A’s current record, I could then run a simulation to determine how often the A’s lose 121 games or more. I chose to simulate 100,000 predicted final winning percentages.
Through 59 games, my model predicted that the A’s have a roughly 16.46% chance to eclipse the 120-loss mark. The mean prediction was a record of 49-113, while the lower and upper bounds were 31-131 and 67-95, respectively, according to a 99% confidence interval.
While this is an interesting experiment, there are several key areas for improvement. First, as Szymborski noted, the A’s will likely look vastly different following the August 1 trade deadline, which in all likelihood will increase their chances of reaching the 121-loss mark. While that type of variation is in some ways baked into the data I collected for each season, the A’s will likely be more aggressive in moving pieces than the average team, and that’s not accounted for directly.
Further, there are several junctures at which I could have taken this project in a different direction. I could have chosen a different cut-off date than 1969, chosen a 95% instead of a 99% confidence interval, and could have run a simulation one thousand or one million times instead of one hundred thousand. Overall, playing with each of these options yielded a similar result -- the A’s have all the pieces (or the lack thereof) to give the ‘62 Mets a run for their money. Each win shaves off a couple percentage points, and each loss adds a tick or two. It’s not by any stretch certain they’ll reach the mark, but it’s nowhere near impossible.
If you’re interested in following the A’s and their chances throughout the rest of the season, I will post figures like the above daily on my Twitter. Thanks for reading.
Works Cited:
Szymborski, D. (2023) Can the Oakland A’s catch the ’62 Mets?, FanGraphs Baseball. Available at: https://blogs.fangraphs.com/can-the-oakland-as-catch-the-62-mets/ (Accessed: 03 June 2023).
Predicting the final record of a baseball team during the season (no date) Apache2 Ubuntu Default Page: It works. Available at: http://spiff.rit.edu/richmond/baseball/record_corr/record_corr.html (Accessed: 03 June 2023).